NRZ, short for Non-Return-to-Zero, is the modulation scheme that built the modern internet. It uses two voltage levels: a high level for a binary 1 and a low level for a binary 0. Each symbol carries exactly one bit.<\/p>\n
The name describes how the signal behaves between bits. The voltage holds its level for the entire bit period instead of returning to zero in between. That simplicity is the entire point.<\/p>\n
<\/p>\n
How NRZ Works<\/strong><\/h3>\nNRZ transmits one bit per symbol. A 25 Gbps lane runs at 25 Gbaud. The receiver only has to tell high from low, which gives it a large margin for error. The result is a single, wide, easy-to-read eye diagram.<\/p>\n
That wide eye is why NRZ is so forgiving. It tolerates noise, reflections, and inexpensive components. NRZ generally allows simpler receiver architectures and may operate without a dedicated DSP or FEC, depending on the application and link requirements.<\/p>\n
<\/p>\n
Where NRZ Is Used<\/strong><\/h3>\nNRZ still dominates the lower half of the optical portfolio:<\/p>\n
\n- 1G and 10G SFP and SFP+ modules<\/li>\n
- 25G SFP28 modules<\/li>\n
- 40G QSFP+ modules typically use four 10G NRZ lanes, while 100G QSFP28 modules commonly use four 25G NRZ lanes.<\/li>\n
- 100G SR4, LR4, CWDM4, and PSM4 optics<\/li>\n<\/ul>\n
<\/p>\n
A 100G QSFP28 SR4 module is, in effect, four 25G NRZ links bundled into one port. That design carried hyperscale data centers through most of the 2010s.\u00a0Some 100G single-lambda transceivers already use PAM4 to achieve higher data rates over a single optical lane.<\/p>\n
<\/p>\n
Where NRZ Hits Its Limit<\/strong><\/h3>\nNRZ has one hard problem: channel bandwidth. To double the data rate, you double the baud rate, which doubles the bandwidth you need. At roughly 25 to 32 Gbps per lane, that bandwidth becomes expensive to move through copper traces, PCB vias, and laser drivers. Loss climbs, and the once-wide eye starts to close.<\/p>\n
That wall is why the industry stopped trying to push NRZ faster and changed strategy entirely. The fix wasn’t a faster clock. It was a smarter encoding.<\/p>\n
<\/p>\n
<\/p>\n
What Is PAM4 Modulation?<\/strong><\/h2>\nPAM4, or 4-level Pulse Amplitude Modulation, is the answer to NRZ’s bandwidth wall. Instead of two voltage levels, it uses four: 00, 01, 10, and 11. Each symbol carries two bits.<\/p>\n
That single change lets PAM4 double the data rate without doubling the baud rate. A 100G lane can run at 50 Gbaud instead of 100 Gbaud. From the other direction, the same baud rate now carries twice as much data.<\/p>\n
<\/p>\n
How PAM4 Works<\/b><\/strong><\/h3>\nFour levels mean the receiver has three decision thresholds instead of one. The eye diagram splits into three smaller eyes stacked on top of each other.<\/p>\n
That stacking is the whole tradeoff. Each PAM4 eye is roughly one-third the height of an NRZ eye at the same voltage swing. The signal is harder to read, which is why PAM4 needs more help downstream.<\/p>\n
<\/p>\n
Why PAM4 Took Over at 400G<\/b><\/strong><\/h3>\nAt 400G, NRZ would need 16 lanes of 25G. PAM4 reaches the same rate with four lanes of 100G or eight lanes of 50G. Fewer lanes mean smaller connectors, fewer fibers, and lower deployment cost. IEEE 802.3bs and 802.3cd standardized PAM4 for exactly this reason.<\/p>\n
Demand has followed. The market for 100G-per-lane PAM4 optics was worth about 2.3\u00a0billion\u00a0in\u00a02024\u00a0and\u00a0is\u00a0projected\u00a0to\u00a0reach\u00a02.3\u00a0billion\u00a0in\u00a02024\u00a0and\u00a0is\u00a0projected\u00a0to\u00a0reach\u00a08.7 billion by 2033, growing at roughly 15.7% a year as AI data centers scale out.<\/p>\n
<\/p>\n
![\"NRZ](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/11\/How-they-work.png\")
<\/p>\n
<\/p>\n
<\/p>\n
PAM4 vs NRZ: Side-by-Side Comparison<\/strong><\/h2>\nThe two schemes differ on far more than how many levels they use. The table below summarizes the parameters that actually matter when you are building a link.<\/p>\n
NRZ still dominates the lower half of the optical portfolio:<\/p>\n
- \n
- 1G and 10G SFP and SFP+ modules<\/li>\n
- 25G SFP28 modules<\/li>\n
- 40G QSFP+ modules typically use four 10G NRZ lanes, while 100G QSFP28 modules commonly use four 25G NRZ lanes.<\/li>\n
- 100G SR4, LR4, CWDM4, and PSM4 optics<\/li>\n<\/ul>\n
<\/p>\n
A 100G QSFP28 SR4 module is, in effect, four 25G NRZ links bundled into one port. That design carried hyperscale data centers through most of the 2010s.\u00a0Some 100G single-lambda transceivers already use PAM4 to achieve higher data rates over a single optical lane.<\/p>\n
<\/p>\n
Where NRZ Hits Its Limit<\/strong><\/h3>\n
NRZ has one hard problem: channel bandwidth. To double the data rate, you double the baud rate, which doubles the bandwidth you need. At roughly 25 to 32 Gbps per lane, that bandwidth becomes expensive to move through copper traces, PCB vias, and laser drivers. Loss climbs, and the once-wide eye starts to close.<\/p>\n
That wall is why the industry stopped trying to push NRZ faster and changed strategy entirely. The fix wasn’t a faster clock. It was a smarter encoding.<\/p>\n
<\/p>\n
<\/p>\n
What Is PAM4 Modulation?<\/strong><\/h2>\n
PAM4, or 4-level Pulse Amplitude Modulation, is the answer to NRZ’s bandwidth wall. Instead of two voltage levels, it uses four: 00, 01, 10, and 11. Each symbol carries two bits.<\/p>\n
That single change lets PAM4 double the data rate without doubling the baud rate. A 100G lane can run at 50 Gbaud instead of 100 Gbaud. From the other direction, the same baud rate now carries twice as much data.<\/p>\n
<\/p>\n
How PAM4 Works<\/b><\/strong><\/h3>\n
Four levels mean the receiver has three decision thresholds instead of one. The eye diagram splits into three smaller eyes stacked on top of each other.<\/p>\n
That stacking is the whole tradeoff. Each PAM4 eye is roughly one-third the height of an NRZ eye at the same voltage swing. The signal is harder to read, which is why PAM4 needs more help downstream.<\/p>\n
<\/p>\n
Why PAM4 Took Over at 400G<\/b><\/strong><\/h3>\n
At 400G, NRZ would need 16 lanes of 25G. PAM4 reaches the same rate with four lanes of 100G or eight lanes of 50G. Fewer lanes mean smaller connectors, fewer fibers, and lower deployment cost. IEEE 802.3bs and 802.3cd standardized PAM4 for exactly this reason.<\/p>\n
Demand has followed. The market for 100G-per-lane PAM4 optics was worth about 2.3\u00a0billion\u00a0in\u00a02024\u00a0and\u00a0is\u00a0projected\u00a0to\u00a0reach\u00a02.3\u00a0billion\u00a0in\u00a02024\u00a0and\u00a0is\u00a0projected\u00a0to\u00a0reach\u00a08.7 billion by 2033, growing at roughly 15.7% a year as AI data centers scale out.<\/p>\n
<\/p>\n
<\/p>\n<\/p>\n
<\/p>\n
PAM4 vs NRZ: Side-by-Side Comparison<\/strong><\/h2>\n
The two schemes differ on far more than how many levels they use. The table below summarizes the parameters that actually matter when you are building a link.<\/p>\n
![\"The](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/11\/PAM4-vs-NRZ.png\")
<\/p>\n![\"From](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/11\/From-100G-NRZ-to-800G-PAM4.png\")
<\/p>\n![\"NRZ](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/11\/NRZ-vs-PAM4-Choosing-the-Right-Modulation-by-Data-Rate.png\")
<\/p>\n